Topology-Based Image Alignment for Repeatable Beam Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing die-to-database (D2DB) alignment techniques for semiconductor inspection images suffer from high variance and poor repeatability, especially in complex device structures like DRAM and FinFETs, due to issues such as off-center shifts and resource-intensive methods, making precise and repeatable alignment challenging.
Innovation Solution
An image alignment method that involves obtaining a reference GDS image, modifying it based on topology characteristics, and aligning inspection images with the modified rendered image to preserve topology, using adaptive filter mechanisms and convolutional matrices to enhance precision and repeatability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing D2DB alignment techniques (edge-based or correlation-based) are used, then throughput is maintained, but alignment precision and repeatability deteriorate due to high variance and poor repeatability in complex device structures
Solution Approach 1:
The patent creates a rendered image that is a synthetic copy of the inspection image by combining database information with process simulation. This rendered image serves as a virtual replica that can be aligned with the actual inspection image, avoiding the high variance issues of direct correlation-based alignment while maintaining sub-nanometer precision through topology-preserving transformations
Solution Approach 2:
The patent performs preliminary rendering of the inspection image using database information and process simulation before actual alignment. By pre-computing the rendered image with expected process variations and topology characteristics, the system establishes a reference that guides subsequent alignment operations, improving both precision and repeatability
2Productivity
If correlation-based alignment techniques are used, then throughput is improved, but alignment reliability deteriorates due to high variance in complex structures like DRAM and FinFETs
Solution Approach 1:
Instead of directly correlating inspection images with database patterns, the patent creates a rendered image copy that incorporates process simulation and topology preservation. This intermediate rendered copy serves as a bridge that maintains alignment reliability while enabling efficient comparison operations
Solution Approach 2:
The rendered image acts as an intermediary between the database information and the inspection image. It mediates the alignment process by incorporating process variations and topology characteristics, reducing the high variance inherent in direct correlation-based methods while maintaining throughput through efficient image processing
3Measurement precision
If traditional image rendering methods are used, then processing speed is maintained, but alignment precision deteriorates due to off-center shifts and failure to preserve topology characteristics
Solution Approach 1:
The patent transforms the rendering approach by changing key parameters: using topology-preserving transformations instead of traditional geometric transformations, and incorporating process simulation parameters to accurately represent process variations. These parameter changes enable sub-nanometer alignment precision while managing complexity through systematic processing steps
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Systems and methods of image alignment are disclosed herein. The method of image alignment may comprise obtaining an image of a sample, obtaining information associated with a corresponding reference image, generating a modified rendered image by blurring a rendered image of the corresponding reference image such that a topology of the rendered image is substantially preserved, wherein a degree of blurring is based on a characteristic of the topology, and aligning the image of the sample with the blurred rendered image. The method may further comprise aligning the image of the sample with the corresponding reference image based on an alignment between the image of the sample and the blurred rendered image.